1 // SPDX-License-Identifier: GPL-2.0-only
3 * This file is part of UBIFS.
5 * Copyright (C) 2006-2008 Nokia Corporation.
7 * Authors: Artem Bityutskiy (Битюцкий Артём)
12 * This file implements UBIFS journal.
14 * The journal consists of 2 parts - the log and bud LEBs. The log has fixed
15 * length and position, while a bud logical eraseblock is any LEB in the main
16 * area. Buds contain file system data - data nodes, inode nodes, etc. The log
17 * contains only references to buds and some other stuff like commit
18 * start node. The idea is that when we commit the journal, we do
19 * not copy the data, the buds just become indexed. Since after the commit the
20 * nodes in bud eraseblocks become leaf nodes of the file system index tree, we
21 * use term "bud". Analogy is obvious, bud eraseblocks contain nodes which will
22 * become leafs in the future.
24 * The journal is multi-headed because we want to write data to the journal as
25 * optimally as possible. It is nice to have nodes belonging to the same inode
26 * in one LEB, so we may write data owned by different inodes to different
27 * journal heads, although at present only one data head is used.
29 * For recovery reasons, the base head contains all inode nodes, all directory
30 * entry nodes and all truncate nodes. This means that the other heads contain
33 * Bud LEBs may be half-indexed. For example, if the bud was not full at the
34 * time of commit, the bud is retained to continue to be used in the journal,
35 * even though the "front" of the LEB is now indexed. In that case, the log
36 * reference contains the offset where the bud starts for the purposes of the
39 * The journal size has to be limited, because the larger is the journal, the
40 * longer it takes to mount UBIFS (scanning the journal) and the more memory it
41 * takes (indexing in the TNC).
43 * All the journal write operations like 'ubifs_jnl_update()' here, which write
44 * multiple UBIFS nodes to the journal at one go, are atomic with respect to
45 * unclean reboots. Should the unclean reboot happen, the recovery code drops
52 * zero_ino_node_unused - zero out unused fields of an on-flash inode node.
53 * @ino: the inode to zero out
55 static inline void zero_ino_node_unused(struct ubifs_ino_node
*ino
)
57 memset(ino
->padding1
, 0, 4);
58 memset(ino
->padding2
, 0, 26);
62 * zero_dent_node_unused - zero out unused fields of an on-flash directory
64 * @dent: the directory entry to zero out
66 static inline void zero_dent_node_unused(struct ubifs_dent_node
*dent
)
72 * zero_trun_node_unused - zero out unused fields of an on-flash truncation
74 * @trun: the truncation node to zero out
76 static inline void zero_trun_node_unused(struct ubifs_trun_node
*trun
)
78 memset(trun
->padding
, 0, 12);
81 static void ubifs_add_auth_dirt(struct ubifs_info
*c
, int lnum
)
83 if (ubifs_authenticated(c
))
84 ubifs_add_dirt(c
, lnum
, ubifs_auth_node_sz(c
));
88 * reserve_space - reserve space in the journal.
89 * @c: UBIFS file-system description object
90 * @jhead: journal head number
93 * This function reserves space in journal head @head. If the reservation
94 * succeeded, the journal head stays locked and later has to be unlocked using
95 * 'release_head()'. Returns zero in case of success, %-EAGAIN if commit has to
96 * be done, and other negative error codes in case of other failures.
98 static int reserve_space(struct ubifs_info
*c
, int jhead
, int len
)
100 int err
= 0, err1
, retries
= 0, avail
, lnum
, offs
, squeeze
;
101 struct ubifs_wbuf
*wbuf
= &c
->jheads
[jhead
].wbuf
;
104 * Typically, the base head has smaller nodes written to it, so it is
105 * better to try to allocate space at the ends of eraseblocks. This is
106 * what the squeeze parameter does.
108 ubifs_assert(c
, !c
->ro_media
&& !c
->ro_mount
);
109 squeeze
= (jhead
== BASEHD
);
111 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
118 avail
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
119 if (wbuf
->lnum
!= -1 && avail
>= len
)
123 * Write buffer wasn't seek'ed or there is no enough space - look for an
124 * LEB with some empty space.
126 lnum
= ubifs_find_free_space(c
, len
, &offs
, squeeze
);
135 * No free space, we have to run garbage collector to make
136 * some. But the write-buffer mutex has to be unlocked because
139 dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead
));
140 mutex_unlock(&wbuf
->io_mutex
);
142 lnum
= ubifs_garbage_collect(c
, 0);
149 * GC could not make a free LEB. But someone else may
150 * have allocated new bud for this journal head,
151 * because we dropped @wbuf->io_mutex, so try once
154 dbg_jnl("GC couldn't make a free LEB for jhead %s",
157 dbg_jnl("retry (%d)", retries
);
161 dbg_jnl("return -ENOSPC");
165 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
166 dbg_jnl("got LEB %d for jhead %s", lnum
, dbg_jhead(jhead
));
167 avail
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
169 if (wbuf
->lnum
!= -1 && avail
>= len
) {
171 * Someone else has switched the journal head and we have
172 * enough space now. This happens when more than one process is
173 * trying to write to the same journal head at the same time.
175 dbg_jnl("return LEB %d back, already have LEB %d:%d",
176 lnum
, wbuf
->lnum
, wbuf
->offs
+ wbuf
->used
);
177 err
= ubifs_return_leb(c
, lnum
);
187 * Make sure we synchronize the write-buffer before we add the new bud
188 * to the log. Otherwise we may have a power cut after the log
189 * reference node for the last bud (@lnum) is written but before the
190 * write-buffer data are written to the next-to-last bud
191 * (@wbuf->lnum). And the effect would be that the recovery would see
192 * that there is corruption in the next-to-last bud.
194 err
= ubifs_wbuf_sync_nolock(wbuf
);
197 err
= ubifs_add_bud_to_log(c
, jhead
, lnum
, offs
);
200 err
= ubifs_wbuf_seek_nolock(wbuf
, lnum
, offs
);
207 mutex_unlock(&wbuf
->io_mutex
);
211 /* An error occurred and the LEB has to be returned to lprops */
212 ubifs_assert(c
, err
< 0);
213 err1
= ubifs_return_leb(c
, lnum
);
214 if (err1
&& err
== -EAGAIN
)
216 * Return original error code only if it is not %-EAGAIN,
217 * which is not really an error. Otherwise, return the error
218 * code of 'ubifs_return_leb()'.
221 mutex_unlock(&wbuf
->io_mutex
);
225 static int ubifs_hash_nodes(struct ubifs_info
*c
, void *node
,
226 int len
, struct shash_desc
*hash
)
228 int auth_node_size
= ubifs_auth_node_sz(c
);
232 const struct ubifs_ch
*ch
= node
;
233 int nodelen
= le32_to_cpu(ch
->len
);
235 ubifs_assert(c
, len
>= auth_node_size
);
237 if (len
== auth_node_size
)
240 ubifs_assert(c
, len
> nodelen
);
241 ubifs_assert(c
, ch
->magic
== cpu_to_le32(UBIFS_NODE_MAGIC
));
243 err
= ubifs_shash_update(c
, hash
, (void *)node
, nodelen
);
247 node
+= ALIGN(nodelen
, 8);
248 len
-= ALIGN(nodelen
, 8);
251 return ubifs_prepare_auth_node(c
, node
, hash
);
255 * write_head - write data to a journal head.
256 * @c: UBIFS file-system description object
257 * @jhead: journal head
258 * @buf: buffer to write
259 * @len: length to write
260 * @lnum: LEB number written is returned here
261 * @offs: offset written is returned here
262 * @sync: non-zero if the write-buffer has to by synchronized
264 * This function writes data to the reserved space of journal head @jhead.
265 * Returns zero in case of success and a negative error code in case of
268 static int write_head(struct ubifs_info
*c
, int jhead
, void *buf
, int len
,
269 int *lnum
, int *offs
, int sync
)
272 struct ubifs_wbuf
*wbuf
= &c
->jheads
[jhead
].wbuf
;
274 ubifs_assert(c
, jhead
!= GCHD
);
276 *lnum
= c
->jheads
[jhead
].wbuf
.lnum
;
277 *offs
= c
->jheads
[jhead
].wbuf
.offs
+ c
->jheads
[jhead
].wbuf
.used
;
278 dbg_jnl("jhead %s, LEB %d:%d, len %d",
279 dbg_jhead(jhead
), *lnum
, *offs
, len
);
281 if (ubifs_authenticated(c
)) {
282 err
= ubifs_hash_nodes(c
, buf
, len
, c
->jheads
[jhead
].log_hash
);
287 err
= ubifs_wbuf_write_nolock(wbuf
, buf
, len
);
291 err
= ubifs_wbuf_sync_nolock(wbuf
);
296 * __queue_and_wait - queue a task and wait until the task is waked up.
297 * @c: UBIFS file-system description object
299 * This function adds current task in queue and waits until the task is waked
300 * up. This function should be called with @c->reserve_space_wq locked.
302 static void __queue_and_wait(struct ubifs_info
*c
)
306 __add_wait_queue_entry_tail_exclusive(&c
->reserve_space_wq
, &wait
);
307 set_current_state(TASK_UNINTERRUPTIBLE
);
308 spin_unlock(&c
->reserve_space_wq
.lock
);
311 finish_wait(&c
->reserve_space_wq
, &wait
);
315 * wait_for_reservation - try queuing current task to wait until waked up.
316 * @c: UBIFS file-system description object
318 * This function queues current task to wait until waked up, if queuing is
319 * started(@c->need_wait_space is not %0). Returns %true if current task is
320 * added in queue, otherwise %false is returned.
322 static bool wait_for_reservation(struct ubifs_info
*c
)
324 if (likely(atomic_read(&c
->need_wait_space
) == 0))
325 /* Quick path to check whether queuing is started. */
328 spin_lock(&c
->reserve_space_wq
.lock
);
329 if (atomic_read(&c
->need_wait_space
) == 0) {
330 /* Queuing is not started, don't queue current task. */
331 spin_unlock(&c
->reserve_space_wq
.lock
);
340 * wake_up_reservation - wake up first task in queue or stop queuing.
341 * @c: UBIFS file-system description object
343 * This function wakes up the first task in queue if it exists, or stops
344 * queuing if no tasks in queue.
346 static void wake_up_reservation(struct ubifs_info
*c
)
348 spin_lock(&c
->reserve_space_wq
.lock
);
349 if (waitqueue_active(&c
->reserve_space_wq
))
350 wake_up_locked(&c
->reserve_space_wq
);
353 * Compared with wait_for_reservation(), set @c->need_wait_space
354 * under the protection of wait queue lock, which can avoid that
355 * @c->need_wait_space is set to 0 after new task queued.
357 atomic_set(&c
->need_wait_space
, 0);
358 spin_unlock(&c
->reserve_space_wq
.lock
);
362 * add_or_start_queue - add current task in queue or start queuing.
363 * @c: UBIFS file-system description object
365 * This function starts queuing if queuing is not started, otherwise adds
366 * current task in queue.
368 static void add_or_start_queue(struct ubifs_info
*c
)
370 spin_lock(&c
->reserve_space_wq
.lock
);
371 if (atomic_cmpxchg(&c
->need_wait_space
, 0, 1) == 0) {
372 /* Starts queuing, task can go on directly. */
373 spin_unlock(&c
->reserve_space_wq
.lock
);
378 * There are at least two tasks have retried more than 32 times
379 * at certain point, first task has started queuing, just queue
386 * make_reservation - reserve journal space.
387 * @c: UBIFS file-system description object
388 * @jhead: journal head
389 * @len: how many bytes to reserve
391 * This function makes space reservation in journal head @jhead. The function
392 * takes the commit lock and locks the journal head, and the caller has to
393 * unlock the head and finish the reservation with 'finish_reservation()'.
394 * Returns zero in case of success and a negative error code in case of
397 * Note, the journal head may be unlocked as soon as the data is written, while
398 * the commit lock has to be released after the data has been added to the
401 static int make_reservation(struct ubifs_info
*c
, int jhead
, int len
)
403 int err
, cmt_retries
= 0, nospc_retries
= 0;
404 bool blocked
= wait_for_reservation(c
);
407 down_read(&c
->commit_sem
);
408 err
= reserve_space(c
, jhead
, len
);
410 /* c->commit_sem will get released via finish_reservation(). */
413 up_read(&c
->commit_sem
);
415 if (err
== -ENOSPC
) {
417 * GC could not make any progress. We should try to commit
418 * because it could make some dirty space and GC would make
419 * progress, so make the error -EAGAIN so that the below
420 * will commit and re-try.
423 dbg_jnl("no space, retry");
431 * -EAGAIN means that the journal is full or too large, or the above
432 * code wants to do one commit. Do this and re-try.
434 if (cmt_retries
> 128) {
436 * This should not happen unless:
437 * 1. The journal size limitations are too tough.
438 * 2. The budgeting is incorrect. We always have to be able to
439 * write to the media, because all operations are budgeted.
440 * Deletions are not budgeted, though, but we reserve an
441 * extra LEB for them.
443 ubifs_err(c
, "stuck in space allocation, nospc_retries %d",
447 } else if (cmt_retries
> 32) {
449 * It's almost impossible to happen, unless there are many tasks
450 * making reservation concurrently and someone task has retried
451 * gc + commit for many times, generated available space during
452 * this period are grabbed by other tasks.
453 * But if it happens, start queuing up all tasks that will make
454 * space reservation, then there is only one task making space
455 * reservation at any time, and it can always make success under
456 * the premise of correct budgeting.
458 ubifs_warn(c
, "too many space allocation cmt_retries (%d) "
459 "nospc_retries (%d), start queuing tasks",
460 cmt_retries
, nospc_retries
);
464 add_or_start_queue(c
);
468 dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
472 err
= ubifs_run_commit(c
);
478 ubifs_err(c
, "cannot reserve %d bytes in jhead %d, error %d",
480 if (err
== -ENOSPC
) {
481 /* This are some budgeting problems, print useful information */
482 down_write(&c
->commit_sem
);
484 ubifs_dump_budg(c
, &c
->bi
);
485 ubifs_dump_lprops(c
);
486 cmt_retries
= dbg_check_lprops(c
);
487 up_write(&c
->commit_sem
);
492 * Only tasks that have ever started queuing or ever been queued
493 * can wake up other queued tasks, which can make sure that
494 * there is only one task waked up to make space reservation.
496 * task A task B task C
497 * make_reservation make_reservation
499 * wake_up_reservation
500 * atomic_cmpxchg // 0, start queuing
502 * wait_for_reservation
505 * if (blocked) // false
506 * // So that task C won't be waked up to race with task B
508 wake_up_reservation(c
);
514 * release_head - release a journal head.
515 * @c: UBIFS file-system description object
516 * @jhead: journal head
518 * This function releases journal head @jhead which was locked by
519 * the 'make_reservation()' function. It has to be called after each successful
520 * 'make_reservation()' invocation.
522 static inline void release_head(struct ubifs_info
*c
, int jhead
)
524 mutex_unlock(&c
->jheads
[jhead
].wbuf
.io_mutex
);
528 * finish_reservation - finish a reservation.
529 * @c: UBIFS file-system description object
531 * This function finishes journal space reservation. It must be called after
532 * 'make_reservation()'.
534 static void finish_reservation(struct ubifs_info
*c
)
536 up_read(&c
->commit_sem
);
540 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
543 static int get_dent_type(int mode
)
545 switch (mode
& S_IFMT
) {
547 return UBIFS_ITYPE_REG
;
549 return UBIFS_ITYPE_DIR
;
551 return UBIFS_ITYPE_LNK
;
553 return UBIFS_ITYPE_BLK
;
555 return UBIFS_ITYPE_CHR
;
557 return UBIFS_ITYPE_FIFO
;
559 return UBIFS_ITYPE_SOCK
;
567 * pack_inode - pack an inode node.
568 * @c: UBIFS file-system description object
569 * @ino: buffer in which to pack inode node
570 * @inode: inode to pack
571 * @last: indicates the last node of the group
573 static void pack_inode(struct ubifs_info
*c
, struct ubifs_ino_node
*ino
,
574 const struct inode
*inode
, int last
)
576 int data_len
= 0, last_reference
= !inode
->i_nlink
;
577 struct ubifs_inode
*ui
= ubifs_inode(inode
);
579 ino
->ch
.node_type
= UBIFS_INO_NODE
;
580 ino_key_init_flash(c
, &ino
->key
, inode
->i_ino
);
581 ino
->creat_sqnum
= cpu_to_le64(ui
->creat_sqnum
);
582 ino
->atime_sec
= cpu_to_le64(inode_get_atime_sec(inode
));
583 ino
->atime_nsec
= cpu_to_le32(inode_get_atime_nsec(inode
));
584 ino
->ctime_sec
= cpu_to_le64(inode_get_ctime_sec(inode
));
585 ino
->ctime_nsec
= cpu_to_le32(inode_get_ctime_nsec(inode
));
586 ino
->mtime_sec
= cpu_to_le64(inode_get_mtime_sec(inode
));
587 ino
->mtime_nsec
= cpu_to_le32(inode_get_mtime_nsec(inode
));
588 ino
->uid
= cpu_to_le32(i_uid_read(inode
));
589 ino
->gid
= cpu_to_le32(i_gid_read(inode
));
590 ino
->mode
= cpu_to_le32(inode
->i_mode
);
591 ino
->flags
= cpu_to_le32(ui
->flags
);
592 ino
->size
= cpu_to_le64(ui
->ui_size
);
593 ino
->nlink
= cpu_to_le32(inode
->i_nlink
);
594 ino
->compr_type
= cpu_to_le16(ui
->compr_type
);
595 ino
->data_len
= cpu_to_le32(ui
->data_len
);
596 ino
->xattr_cnt
= cpu_to_le32(ui
->xattr_cnt
);
597 ino
->xattr_size
= cpu_to_le32(ui
->xattr_size
);
598 ino
->xattr_names
= cpu_to_le32(ui
->xattr_names
);
599 zero_ino_node_unused(ino
);
602 * Drop the attached data if this is a deletion inode, the data is not
605 if (!last_reference
) {
606 memcpy(ino
->data
, ui
->data
, ui
->data_len
);
607 data_len
= ui
->data_len
;
610 ubifs_prep_grp_node(c
, ino
, UBIFS_INO_NODE_SZ
+ data_len
, last
);
614 * mark_inode_clean - mark UBIFS inode as clean.
615 * @c: UBIFS file-system description object
616 * @ui: UBIFS inode to mark as clean
618 * This helper function marks UBIFS inode @ui as clean by cleaning the
619 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
620 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
623 static void mark_inode_clean(struct ubifs_info
*c
, struct ubifs_inode
*ui
)
626 ubifs_release_dirty_inode_budget(c
, ui
);
630 static void set_dent_cookie(struct ubifs_info
*c
, struct ubifs_dent_node
*dent
)
633 dent
->cookie
= (__force __le32
) get_random_u32();
639 * ubifs_jnl_update - update inode.
640 * @c: UBIFS file-system description object
641 * @dir: parent inode or host inode in case of extended attributes
642 * @nm: directory entry name
643 * @inode: inode to update
644 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
645 * @xent: non-zero if the directory entry is an extended attribute entry
646 * @in_orphan: indicates whether the @inode is in orphan list
648 * This function updates an inode by writing a directory entry (or extended
649 * attribute entry), the inode itself, and the parent directory inode (or the
650 * host inode) to the journal.
652 * The function writes the host inode @dir last, which is important in case of
653 * extended attributes. Indeed, then we guarantee that if the host inode gets
654 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
655 * the extended attribute inode gets flushed too. And this is exactly what the
656 * user expects - synchronizing the host inode synchronizes its extended
657 * attributes. Similarly, this guarantees that if @dir is synchronized, its
658 * directory entry corresponding to @nm gets synchronized too.
660 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
661 * function synchronizes the write-buffer.
663 * This function marks the @dir and @inode inodes as clean and returns zero on
664 * success. In case of failure, a negative error code is returned.
666 int ubifs_jnl_update(struct ubifs_info
*c
, const struct inode
*dir
,
667 const struct fscrypt_name
*nm
, const struct inode
*inode
,
668 int deletion
, int xent
, int in_orphan
)
670 int err
, dlen
, ilen
, len
, lnum
, ino_offs
, dent_offs
, orphan_added
= 0;
671 int aligned_dlen
, aligned_ilen
, sync
= IS_DIRSYNC(dir
);
672 int last_reference
= !!(deletion
&& inode
->i_nlink
== 0);
673 struct ubifs_inode
*ui
= ubifs_inode(inode
);
674 struct ubifs_inode
*host_ui
= ubifs_inode(dir
);
675 struct ubifs_dent_node
*dent
;
676 struct ubifs_ino_node
*ino
;
677 union ubifs_key dent_key
, ino_key
;
678 u8 hash_dent
[UBIFS_HASH_ARR_SZ
];
679 u8 hash_ino
[UBIFS_HASH_ARR_SZ
];
680 u8 hash_ino_host
[UBIFS_HASH_ARR_SZ
];
682 ubifs_assert(c
, mutex_is_locked(&host_ui
->ui_mutex
));
684 dlen
= UBIFS_DENT_NODE_SZ
+ fname_len(nm
) + 1;
685 ilen
= UBIFS_INO_NODE_SZ
;
688 * If the last reference to the inode is being deleted, then there is
689 * no need to attach and write inode data, it is being deleted anyway.
690 * And if the inode is being deleted, no need to synchronize
691 * write-buffer even if the inode is synchronous.
693 if (!last_reference
) {
694 ilen
+= ui
->data_len
;
695 sync
|= IS_SYNC(inode
);
698 aligned_dlen
= ALIGN(dlen
, 8);
699 aligned_ilen
= ALIGN(ilen
, 8);
701 len
= aligned_dlen
+ aligned_ilen
+ UBIFS_INO_NODE_SZ
;
702 /* Make sure to also account for extended attributes */
703 if (ubifs_authenticated(c
))
704 len
+= ALIGN(host_ui
->data_len
, 8) + ubifs_auth_node_sz(c
);
706 len
+= host_ui
->data_len
;
708 dent
= kzalloc(len
, GFP_NOFS
);
712 /* Make reservation before allocating sequence numbers */
713 err
= make_reservation(c
, BASEHD
, len
);
718 dent
->ch
.node_type
= UBIFS_DENT_NODE
;
719 if (fname_name(nm
) == NULL
)
720 dent_key_init_hash(c
, &dent_key
, dir
->i_ino
, nm
->hash
);
722 dent_key_init(c
, &dent_key
, dir
->i_ino
, nm
);
724 dent
->ch
.node_type
= UBIFS_XENT_NODE
;
725 xent_key_init(c
, &dent_key
, dir
->i_ino
, nm
);
728 key_write(c
, &dent_key
, dent
->key
);
729 dent
->inum
= deletion
? 0 : cpu_to_le64(inode
->i_ino
);
730 dent
->type
= get_dent_type(inode
->i_mode
);
731 dent
->nlen
= cpu_to_le16(fname_len(nm
));
732 memcpy(dent
->name
, fname_name(nm
), fname_len(nm
));
733 dent
->name
[fname_len(nm
)] = '\0';
734 set_dent_cookie(c
, dent
);
736 zero_dent_node_unused(dent
);
737 ubifs_prep_grp_node(c
, dent
, dlen
, 0);
738 err
= ubifs_node_calc_hash(c
, dent
, hash_dent
);
742 ino
= (void *)dent
+ aligned_dlen
;
743 pack_inode(c
, ino
, inode
, 0);
744 err
= ubifs_node_calc_hash(c
, ino
, hash_ino
);
748 ino
= (void *)ino
+ aligned_ilen
;
749 pack_inode(c
, ino
, dir
, 1);
750 err
= ubifs_node_calc_hash(c
, ino
, hash_ino_host
);
754 if (last_reference
&& !in_orphan
) {
755 err
= ubifs_add_orphan(c
, inode
->i_ino
);
757 release_head(c
, BASEHD
);
760 ui
->del_cmtno
= c
->cmt_no
;
764 err
= write_head(c
, BASEHD
, dent
, len
, &lnum
, &dent_offs
, sync
);
768 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
770 ubifs_wbuf_add_ino_nolock(wbuf
, inode
->i_ino
);
771 ubifs_wbuf_add_ino_nolock(wbuf
, dir
->i_ino
);
773 release_head(c
, BASEHD
);
775 ubifs_add_auth_dirt(c
, lnum
);
778 if (fname_name(nm
) == NULL
)
779 err
= ubifs_tnc_remove_dh(c
, &dent_key
, nm
->minor_hash
);
781 err
= ubifs_tnc_remove_nm(c
, &dent_key
, nm
);
784 err
= ubifs_add_dirt(c
, lnum
, dlen
);
786 err
= ubifs_tnc_add_nm(c
, &dent_key
, lnum
, dent_offs
, dlen
,
792 * Note, we do not remove the inode from TNC even if the last reference
793 * to it has just been deleted, because the inode may still be opened.
794 * Instead, the inode has been added to orphan lists and the orphan
795 * subsystem will take further care about it.
797 ino_key_init(c
, &ino_key
, inode
->i_ino
);
798 ino_offs
= dent_offs
+ aligned_dlen
;
799 err
= ubifs_tnc_add(c
, &ino_key
, lnum
, ino_offs
, ilen
, hash_ino
);
803 ino_key_init(c
, &ino_key
, dir
->i_ino
);
804 ino_offs
+= aligned_ilen
;
805 err
= ubifs_tnc_add(c
, &ino_key
, lnum
, ino_offs
,
806 UBIFS_INO_NODE_SZ
+ host_ui
->data_len
, hash_ino_host
);
810 if (in_orphan
&& inode
->i_nlink
)
811 ubifs_delete_orphan(c
, inode
->i_ino
);
813 finish_reservation(c
);
814 spin_lock(&ui
->ui_lock
);
815 ui
->synced_i_size
= ui
->ui_size
;
816 spin_unlock(&ui
->ui_lock
);
818 spin_lock(&host_ui
->ui_lock
);
819 host_ui
->synced_i_size
= host_ui
->ui_size
;
820 spin_unlock(&host_ui
->ui_lock
);
822 mark_inode_clean(c
, ui
);
823 mark_inode_clean(c
, host_ui
);
827 finish_reservation(c
);
833 release_head(c
, BASEHD
);
836 ubifs_ro_mode(c
, err
);
838 ubifs_delete_orphan(c
, inode
->i_ino
);
839 finish_reservation(c
);
844 * ubifs_jnl_write_data - write a data node to the journal.
845 * @c: UBIFS file-system description object
846 * @inode: inode the data node belongs to
848 * @buf: buffer to write
849 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
851 * This function writes a data node to the journal. Returns %0 if the data node
852 * was successfully written, and a negative error code in case of failure.
854 int ubifs_jnl_write_data(struct ubifs_info
*c
, const struct inode
*inode
,
855 const union ubifs_key
*key
, const void *buf
, int len
)
857 struct ubifs_data_node
*data
;
858 int err
, lnum
, offs
, compr_type
, out_len
, compr_len
, auth_len
;
859 int dlen
= COMPRESSED_DATA_NODE_BUF_SZ
, allocated
= 1;
861 struct ubifs_inode
*ui
= ubifs_inode(inode
);
862 bool encrypted
= IS_ENCRYPTED(inode
);
863 u8 hash
[UBIFS_HASH_ARR_SZ
];
865 dbg_jnlk(key
, "ino %lu, blk %u, len %d, key ",
866 (unsigned long)key_inum(c
, key
), key_block(c
, key
), len
);
867 ubifs_assert(c
, len
<= UBIFS_BLOCK_SIZE
);
870 dlen
+= UBIFS_CIPHER_BLOCK_SIZE
;
872 auth_len
= ubifs_auth_node_sz(c
);
874 data
= kmalloc(dlen
+ auth_len
, GFP_NOFS
| __GFP_NOWARN
);
877 * Fall-back to the write reserve buffer. Note, we might be
878 * currently on the memory reclaim path, when the kernel is
879 * trying to free some memory by writing out dirty pages. The
880 * write reserve buffer helps us to guarantee that we are
881 * always able to write the data.
884 mutex_lock(&c
->write_reserve_mutex
);
885 data
= c
->write_reserve_buf
;
888 data
->ch
.node_type
= UBIFS_DATA_NODE
;
889 key_write(c
, key
, &data
->key
);
890 data
->size
= cpu_to_le32(len
);
892 if (!(ui
->flags
& UBIFS_COMPR_FL
))
893 /* Compression is disabled for this inode */
894 compr_type
= UBIFS_COMPR_NONE
;
896 compr_type
= ui
->compr_type
;
898 out_len
= compr_len
= dlen
- UBIFS_DATA_NODE_SZ
;
899 ubifs_compress(c
, buf
, len
, &data
->data
, &compr_len
, &compr_type
);
900 ubifs_assert(c
, compr_len
<= UBIFS_BLOCK_SIZE
);
903 err
= ubifs_encrypt(inode
, data
, compr_len
, &out_len
, key_block(c
, key
));
908 data
->compr_size
= 0;
912 dlen
= UBIFS_DATA_NODE_SZ
+ out_len
;
913 if (ubifs_authenticated(c
))
914 write_len
= ALIGN(dlen
, 8) + auth_len
;
918 data
->compr_type
= cpu_to_le16(compr_type
);
920 /* Make reservation before allocating sequence numbers */
921 err
= make_reservation(c
, DATAHD
, write_len
);
925 ubifs_prepare_node(c
, data
, dlen
, 0);
926 err
= write_head(c
, DATAHD
, data
, write_len
, &lnum
, &offs
, 0);
930 err
= ubifs_node_calc_hash(c
, data
, hash
);
934 ubifs_wbuf_add_ino_nolock(&c
->jheads
[DATAHD
].wbuf
, key_inum(c
, key
));
935 release_head(c
, DATAHD
);
937 ubifs_add_auth_dirt(c
, lnum
);
939 err
= ubifs_tnc_add(c
, key
, lnum
, offs
, dlen
, hash
);
943 finish_reservation(c
);
945 mutex_unlock(&c
->write_reserve_mutex
);
951 release_head(c
, DATAHD
);
953 ubifs_ro_mode(c
, err
);
954 finish_reservation(c
);
957 mutex_unlock(&c
->write_reserve_mutex
);
964 * ubifs_jnl_write_inode - flush inode to the journal.
965 * @c: UBIFS file-system description object
966 * @inode: inode to flush
968 * This function writes inode @inode to the journal. If the inode is
969 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
970 * success and a negative error code in case of failure.
972 int ubifs_jnl_write_inode(struct ubifs_info
*c
, const struct inode
*inode
)
975 struct ubifs_ino_node
*ino
, *ino_start
;
976 struct ubifs_inode
*ui
= ubifs_inode(inode
);
977 int sync
= 0, write_len
= 0, ilen
= UBIFS_INO_NODE_SZ
;
978 int last_reference
= !inode
->i_nlink
;
979 int kill_xattrs
= ui
->xattr_cnt
&& last_reference
;
980 u8 hash
[UBIFS_HASH_ARR_SZ
];
982 dbg_jnl("ino %lu, nlink %u", inode
->i_ino
, inode
->i_nlink
);
985 * If the inode is being deleted, do not write the attached data. No
986 * need to synchronize the write-buffer either.
988 if (!last_reference
) {
989 ilen
+= ui
->data_len
;
990 sync
= IS_SYNC(inode
);
991 } else if (kill_xattrs
) {
992 write_len
+= UBIFS_INO_NODE_SZ
* ui
->xattr_cnt
;
995 if (ubifs_authenticated(c
))
996 write_len
+= ALIGN(ilen
, 8) + ubifs_auth_node_sz(c
);
1000 ino_start
= ino
= kmalloc(write_len
, GFP_NOFS
);
1004 /* Make reservation before allocating sequence numbers */
1005 err
= make_reservation(c
, BASEHD
, write_len
);
1010 union ubifs_key key
;
1011 struct fscrypt_name nm
= {0};
1013 struct ubifs_dent_node
*xent
, *pxent
= NULL
;
1015 if (ui
->xattr_cnt
> ubifs_xattr_max_cnt(c
)) {
1017 ubifs_err(c
, "Cannot delete inode, it has too much xattrs!");
1021 lowest_xent_key(c
, &key
, inode
->i_ino
);
1023 xent
= ubifs_tnc_next_ent(c
, &key
, &nm
);
1025 err
= PTR_ERR(xent
);
1033 fname_name(&nm
) = xent
->name
;
1034 fname_len(&nm
) = le16_to_cpu(xent
->nlen
);
1036 xino
= ubifs_iget(c
->vfs_sb
, le64_to_cpu(xent
->inum
));
1038 err
= PTR_ERR(xino
);
1039 ubifs_err(c
, "dead directory entry '%s', error %d",
1041 ubifs_ro_mode(c
, err
);
1046 ubifs_assert(c
, ubifs_inode(xino
)->xattr
);
1049 pack_inode(c
, ino
, xino
, 0);
1050 ino
= (void *)ino
+ UBIFS_INO_NODE_SZ
;
1055 key_read(c
, &xent
->key
, &key
);
1060 pack_inode(c
, ino
, inode
, 1);
1061 err
= ubifs_node_calc_hash(c
, ino
, hash
);
1065 err
= write_head(c
, BASEHD
, ino_start
, write_len
, &lnum
, &offs
, sync
);
1069 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
,
1071 release_head(c
, BASEHD
);
1073 if (last_reference
) {
1074 err
= ubifs_tnc_remove_ino(c
, inode
->i_ino
);
1077 ubifs_delete_orphan(c
, inode
->i_ino
);
1078 err
= ubifs_add_dirt(c
, lnum
, write_len
);
1080 union ubifs_key key
;
1082 ubifs_add_auth_dirt(c
, lnum
);
1084 ino_key_init(c
, &key
, inode
->i_ino
);
1085 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, ilen
, hash
);
1090 finish_reservation(c
);
1091 spin_lock(&ui
->ui_lock
);
1092 ui
->synced_i_size
= ui
->ui_size
;
1093 spin_unlock(&ui
->ui_lock
);
1098 release_head(c
, BASEHD
);
1100 ubifs_ro_mode(c
, err
);
1101 finish_reservation(c
);
1108 * ubifs_jnl_delete_inode - delete an inode.
1109 * @c: UBIFS file-system description object
1110 * @inode: inode to delete
1112 * This function deletes inode @inode which includes removing it from orphans,
1113 * deleting it from TNC and, in some cases, writing a deletion inode to the
1116 * When regular file inodes are unlinked or a directory inode is removed, the
1117 * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
1118 * direntry to the media, and adds the inode to orphans. After this, when the
1119 * last reference to this inode has been dropped, this function is called. In
1120 * general, it has to write one more deletion inode to the media, because if
1121 * a commit happened between 'ubifs_jnl_update()' and
1122 * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
1123 * anymore, and in fact it might not be on the flash anymore, because it might
1124 * have been garbage-collected already. And for optimization reasons UBIFS does
1125 * not read the orphan area if it has been unmounted cleanly, so it would have
1126 * no indication in the journal that there is a deleted inode which has to be
1129 * However, if there was no commit between 'ubifs_jnl_update()' and
1130 * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
1131 * inode to the media for the second time. And this is quite a typical case.
1133 * This function returns zero in case of success and a negative error code in
1136 int ubifs_jnl_delete_inode(struct ubifs_info
*c
, const struct inode
*inode
)
1139 struct ubifs_inode
*ui
= ubifs_inode(inode
);
1141 ubifs_assert(c
, inode
->i_nlink
== 0);
1143 if (ui
->xattr_cnt
|| ui
->del_cmtno
!= c
->cmt_no
)
1144 /* A commit happened for sure or inode hosts xattrs */
1145 return ubifs_jnl_write_inode(c
, inode
);
1147 down_read(&c
->commit_sem
);
1149 * Check commit number again, because the first test has been done
1150 * without @c->commit_sem, so a commit might have happened.
1152 if (ui
->del_cmtno
!= c
->cmt_no
) {
1153 up_read(&c
->commit_sem
);
1154 return ubifs_jnl_write_inode(c
, inode
);
1157 err
= ubifs_tnc_remove_ino(c
, inode
->i_ino
);
1159 ubifs_ro_mode(c
, err
);
1161 ubifs_delete_orphan(c
, inode
->i_ino
);
1162 up_read(&c
->commit_sem
);
1167 * ubifs_jnl_xrename - cross rename two directory entries.
1168 * @c: UBIFS file-system description object
1169 * @fst_dir: parent inode of 1st directory entry to exchange
1170 * @fst_inode: 1st inode to exchange
1171 * @fst_nm: name of 1st inode to exchange
1172 * @snd_dir: parent inode of 2nd directory entry to exchange
1173 * @snd_inode: 2nd inode to exchange
1174 * @snd_nm: name of 2nd inode to exchange
1175 * @sync: non-zero if the write-buffer has to be synchronized
1177 * This function implements the cross rename operation which may involve
1178 * writing 2 inodes and 2 directory entries. It marks the written inodes as clean
1179 * and returns zero on success. In case of failure, a negative error code is
1182 int ubifs_jnl_xrename(struct ubifs_info
*c
, const struct inode
*fst_dir
,
1183 const struct inode
*fst_inode
,
1184 const struct fscrypt_name
*fst_nm
,
1185 const struct inode
*snd_dir
,
1186 const struct inode
*snd_inode
,
1187 const struct fscrypt_name
*snd_nm
, int sync
)
1189 union ubifs_key key
;
1190 struct ubifs_dent_node
*dent1
, *dent2
;
1191 int err
, dlen1
, dlen2
, lnum
, offs
, len
, plen
= UBIFS_INO_NODE_SZ
;
1192 int aligned_dlen1
, aligned_dlen2
;
1193 int twoparents
= (fst_dir
!= snd_dir
);
1195 u8 hash_dent1
[UBIFS_HASH_ARR_SZ
];
1196 u8 hash_dent2
[UBIFS_HASH_ARR_SZ
];
1197 u8 hash_p1
[UBIFS_HASH_ARR_SZ
];
1198 u8 hash_p2
[UBIFS_HASH_ARR_SZ
];
1200 ubifs_assert(c
, ubifs_inode(fst_dir
)->data_len
== 0);
1201 ubifs_assert(c
, ubifs_inode(snd_dir
)->data_len
== 0);
1202 ubifs_assert(c
, mutex_is_locked(&ubifs_inode(fst_dir
)->ui_mutex
));
1203 ubifs_assert(c
, mutex_is_locked(&ubifs_inode(snd_dir
)->ui_mutex
));
1205 dlen1
= UBIFS_DENT_NODE_SZ
+ fname_len(snd_nm
) + 1;
1206 dlen2
= UBIFS_DENT_NODE_SZ
+ fname_len(fst_nm
) + 1;
1207 aligned_dlen1
= ALIGN(dlen1
, 8);
1208 aligned_dlen2
= ALIGN(dlen2
, 8);
1210 len
= aligned_dlen1
+ aligned_dlen2
+ ALIGN(plen
, 8);
1214 len
+= ubifs_auth_node_sz(c
);
1216 dent1
= kzalloc(len
, GFP_NOFS
);
1220 /* Make reservation before allocating sequence numbers */
1221 err
= make_reservation(c
, BASEHD
, len
);
1225 /* Make new dent for 1st entry */
1226 dent1
->ch
.node_type
= UBIFS_DENT_NODE
;
1227 dent_key_init_flash(c
, &dent1
->key
, snd_dir
->i_ino
, snd_nm
);
1228 dent1
->inum
= cpu_to_le64(fst_inode
->i_ino
);
1229 dent1
->type
= get_dent_type(fst_inode
->i_mode
);
1230 dent1
->nlen
= cpu_to_le16(fname_len(snd_nm
));
1231 memcpy(dent1
->name
, fname_name(snd_nm
), fname_len(snd_nm
));
1232 dent1
->name
[fname_len(snd_nm
)] = '\0';
1233 set_dent_cookie(c
, dent1
);
1234 zero_dent_node_unused(dent1
);
1235 ubifs_prep_grp_node(c
, dent1
, dlen1
, 0);
1236 err
= ubifs_node_calc_hash(c
, dent1
, hash_dent1
);
1240 /* Make new dent for 2nd entry */
1241 dent2
= (void *)dent1
+ aligned_dlen1
;
1242 dent2
->ch
.node_type
= UBIFS_DENT_NODE
;
1243 dent_key_init_flash(c
, &dent2
->key
, fst_dir
->i_ino
, fst_nm
);
1244 dent2
->inum
= cpu_to_le64(snd_inode
->i_ino
);
1245 dent2
->type
= get_dent_type(snd_inode
->i_mode
);
1246 dent2
->nlen
= cpu_to_le16(fname_len(fst_nm
));
1247 memcpy(dent2
->name
, fname_name(fst_nm
), fname_len(fst_nm
));
1248 dent2
->name
[fname_len(fst_nm
)] = '\0';
1249 set_dent_cookie(c
, dent2
);
1250 zero_dent_node_unused(dent2
);
1251 ubifs_prep_grp_node(c
, dent2
, dlen2
, 0);
1252 err
= ubifs_node_calc_hash(c
, dent2
, hash_dent2
);
1256 p
= (void *)dent2
+ aligned_dlen2
;
1258 pack_inode(c
, p
, fst_dir
, 1);
1259 err
= ubifs_node_calc_hash(c
, p
, hash_p1
);
1263 pack_inode(c
, p
, fst_dir
, 0);
1264 err
= ubifs_node_calc_hash(c
, p
, hash_p1
);
1267 p
+= ALIGN(plen
, 8);
1268 pack_inode(c
, p
, snd_dir
, 1);
1269 err
= ubifs_node_calc_hash(c
, p
, hash_p2
);
1274 err
= write_head(c
, BASEHD
, dent1
, len
, &lnum
, &offs
, sync
);
1278 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
1280 ubifs_wbuf_add_ino_nolock(wbuf
, fst_dir
->i_ino
);
1281 ubifs_wbuf_add_ino_nolock(wbuf
, snd_dir
->i_ino
);
1283 release_head(c
, BASEHD
);
1285 ubifs_add_auth_dirt(c
, lnum
);
1287 dent_key_init(c
, &key
, snd_dir
->i_ino
, snd_nm
);
1288 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen1
, hash_dent1
, snd_nm
);
1292 offs
+= aligned_dlen1
;
1293 dent_key_init(c
, &key
, fst_dir
->i_ino
, fst_nm
);
1294 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen2
, hash_dent2
, fst_nm
);
1298 offs
+= aligned_dlen2
;
1300 ino_key_init(c
, &key
, fst_dir
->i_ino
);
1301 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
, hash_p1
);
1306 offs
+= ALIGN(plen
, 8);
1307 ino_key_init(c
, &key
, snd_dir
->i_ino
);
1308 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
, hash_p2
);
1313 finish_reservation(c
);
1315 mark_inode_clean(c
, ubifs_inode(fst_dir
));
1317 mark_inode_clean(c
, ubifs_inode(snd_dir
));
1322 release_head(c
, BASEHD
);
1324 ubifs_ro_mode(c
, err
);
1325 finish_reservation(c
);
1332 * ubifs_jnl_rename - rename a directory entry.
1333 * @c: UBIFS file-system description object
1334 * @old_dir: parent inode of directory entry to rename
1335 * @old_inode: directory entry's inode to rename
1336 * @old_nm: name of the old directory entry to rename
1337 * @new_dir: parent inode of directory entry to rename
1338 * @new_inode: new directory entry's inode (or directory entry's inode to
1340 * @new_nm: new name of the new directory entry
1341 * @whiteout: whiteout inode
1342 * @sync: non-zero if the write-buffer has to be synchronized
1343 * @delete_orphan: indicates an orphan entry deletion for @whiteout
1345 * This function implements the re-name operation which may involve writing up
1346 * to 4 inodes(new inode, whiteout inode, old and new parent directory inodes)
1347 * and 2 directory entries. It marks the written inodes as clean and returns
1348 * zero on success. In case of failure, a negative error code is returned.
1350 int ubifs_jnl_rename(struct ubifs_info
*c
, const struct inode
*old_dir
,
1351 const struct inode
*old_inode
,
1352 const struct fscrypt_name
*old_nm
,
1353 const struct inode
*new_dir
,
1354 const struct inode
*new_inode
,
1355 const struct fscrypt_name
*new_nm
,
1356 const struct inode
*whiteout
, int sync
, int delete_orphan
)
1359 union ubifs_key key
;
1360 struct ubifs_dent_node
*dent
, *dent2
;
1361 int err
, dlen1
, dlen2
, ilen
, wlen
, lnum
, offs
, len
, orphan_added
= 0;
1362 int aligned_dlen1
, aligned_dlen2
, plen
= UBIFS_INO_NODE_SZ
;
1363 int last_reference
= !!(new_inode
&& new_inode
->i_nlink
== 0);
1364 int move
= (old_dir
!= new_dir
);
1365 struct ubifs_inode
*new_ui
, *whiteout_ui
;
1366 u8 hash_old_dir
[UBIFS_HASH_ARR_SZ
];
1367 u8 hash_new_dir
[UBIFS_HASH_ARR_SZ
];
1368 u8 hash_new_inode
[UBIFS_HASH_ARR_SZ
];
1369 u8 hash_whiteout_inode
[UBIFS_HASH_ARR_SZ
];
1370 u8 hash_dent1
[UBIFS_HASH_ARR_SZ
];
1371 u8 hash_dent2
[UBIFS_HASH_ARR_SZ
];
1373 ubifs_assert(c
, ubifs_inode(old_dir
)->data_len
== 0);
1374 ubifs_assert(c
, ubifs_inode(new_dir
)->data_len
== 0);
1375 ubifs_assert(c
, mutex_is_locked(&ubifs_inode(old_dir
)->ui_mutex
));
1376 ubifs_assert(c
, mutex_is_locked(&ubifs_inode(new_dir
)->ui_mutex
));
1378 dlen1
= UBIFS_DENT_NODE_SZ
+ fname_len(new_nm
) + 1;
1379 dlen2
= UBIFS_DENT_NODE_SZ
+ fname_len(old_nm
) + 1;
1381 new_ui
= ubifs_inode(new_inode
);
1382 ubifs_assert(c
, mutex_is_locked(&new_ui
->ui_mutex
));
1383 ilen
= UBIFS_INO_NODE_SZ
;
1384 if (!last_reference
)
1385 ilen
+= new_ui
->data_len
;
1390 whiteout_ui
= ubifs_inode(whiteout
);
1391 ubifs_assert(c
, mutex_is_locked(&whiteout_ui
->ui_mutex
));
1392 ubifs_assert(c
, whiteout
->i_nlink
== 1);
1393 ubifs_assert(c
, !whiteout_ui
->dirty
);
1394 wlen
= UBIFS_INO_NODE_SZ
;
1395 wlen
+= whiteout_ui
->data_len
;
1399 aligned_dlen1
= ALIGN(dlen1
, 8);
1400 aligned_dlen2
= ALIGN(dlen2
, 8);
1401 len
= aligned_dlen1
+ aligned_dlen2
+ ALIGN(ilen
, 8) +
1402 ALIGN(wlen
, 8) + ALIGN(plen
, 8);
1406 len
+= ubifs_auth_node_sz(c
);
1408 dent
= kzalloc(len
, GFP_NOFS
);
1412 /* Make reservation before allocating sequence numbers */
1413 err
= make_reservation(c
, BASEHD
, len
);
1418 dent
->ch
.node_type
= UBIFS_DENT_NODE
;
1419 dent_key_init_flash(c
, &dent
->key
, new_dir
->i_ino
, new_nm
);
1420 dent
->inum
= cpu_to_le64(old_inode
->i_ino
);
1421 dent
->type
= get_dent_type(old_inode
->i_mode
);
1422 dent
->nlen
= cpu_to_le16(fname_len(new_nm
));
1423 memcpy(dent
->name
, fname_name(new_nm
), fname_len(new_nm
));
1424 dent
->name
[fname_len(new_nm
)] = '\0';
1425 set_dent_cookie(c
, dent
);
1426 zero_dent_node_unused(dent
);
1427 ubifs_prep_grp_node(c
, dent
, dlen1
, 0);
1428 err
= ubifs_node_calc_hash(c
, dent
, hash_dent1
);
1432 dent2
= (void *)dent
+ aligned_dlen1
;
1433 dent2
->ch
.node_type
= UBIFS_DENT_NODE
;
1434 dent_key_init_flash(c
, &dent2
->key
, old_dir
->i_ino
, old_nm
);
1437 dent2
->inum
= cpu_to_le64(whiteout
->i_ino
);
1438 dent2
->type
= get_dent_type(whiteout
->i_mode
);
1440 /* Make deletion dent */
1442 dent2
->type
= DT_UNKNOWN
;
1444 dent2
->nlen
= cpu_to_le16(fname_len(old_nm
));
1445 memcpy(dent2
->name
, fname_name(old_nm
), fname_len(old_nm
));
1446 dent2
->name
[fname_len(old_nm
)] = '\0';
1447 set_dent_cookie(c
, dent2
);
1448 zero_dent_node_unused(dent2
);
1449 ubifs_prep_grp_node(c
, dent2
, dlen2
, 0);
1450 err
= ubifs_node_calc_hash(c
, dent2
, hash_dent2
);
1454 p
= (void *)dent2
+ aligned_dlen2
;
1456 pack_inode(c
, p
, new_inode
, 0);
1457 err
= ubifs_node_calc_hash(c
, p
, hash_new_inode
);
1461 p
+= ALIGN(ilen
, 8);
1465 pack_inode(c
, p
, whiteout
, 0);
1466 err
= ubifs_node_calc_hash(c
, p
, hash_whiteout_inode
);
1470 p
+= ALIGN(wlen
, 8);
1474 pack_inode(c
, p
, old_dir
, 1);
1475 err
= ubifs_node_calc_hash(c
, p
, hash_old_dir
);
1479 pack_inode(c
, p
, old_dir
, 0);
1480 err
= ubifs_node_calc_hash(c
, p
, hash_old_dir
);
1484 p
+= ALIGN(plen
, 8);
1485 pack_inode(c
, p
, new_dir
, 1);
1486 err
= ubifs_node_calc_hash(c
, p
, hash_new_dir
);
1491 if (last_reference
) {
1492 err
= ubifs_add_orphan(c
, new_inode
->i_ino
);
1494 release_head(c
, BASEHD
);
1497 new_ui
->del_cmtno
= c
->cmt_no
;
1501 err
= write_head(c
, BASEHD
, dent
, len
, &lnum
, &offs
, sync
);
1505 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
1507 ubifs_wbuf_add_ino_nolock(wbuf
, new_dir
->i_ino
);
1508 ubifs_wbuf_add_ino_nolock(wbuf
, old_dir
->i_ino
);
1510 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
,
1513 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
,
1516 release_head(c
, BASEHD
);
1518 ubifs_add_auth_dirt(c
, lnum
);
1520 dent_key_init(c
, &key
, new_dir
->i_ino
, new_nm
);
1521 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen1
, hash_dent1
, new_nm
);
1525 offs
+= aligned_dlen1
;
1527 dent_key_init(c
, &key
, old_dir
->i_ino
, old_nm
);
1528 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen2
, hash_dent2
, old_nm
);
1532 err
= ubifs_add_dirt(c
, lnum
, dlen2
);
1536 dent_key_init(c
, &key
, old_dir
->i_ino
, old_nm
);
1537 err
= ubifs_tnc_remove_nm(c
, &key
, old_nm
);
1542 offs
+= aligned_dlen2
;
1544 ino_key_init(c
, &key
, new_inode
->i_ino
);
1545 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, ilen
, hash_new_inode
);
1548 offs
+= ALIGN(ilen
, 8);
1552 ino_key_init(c
, &key
, whiteout
->i_ino
);
1553 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, wlen
,
1554 hash_whiteout_inode
);
1557 offs
+= ALIGN(wlen
, 8);
1560 ino_key_init(c
, &key
, old_dir
->i_ino
);
1561 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
, hash_old_dir
);
1566 offs
+= ALIGN(plen
, 8);
1567 ino_key_init(c
, &key
, new_dir
->i_ino
);
1568 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
, hash_new_dir
);
1574 ubifs_delete_orphan(c
, whiteout
->i_ino
);
1576 finish_reservation(c
);
1578 mark_inode_clean(c
, new_ui
);
1579 spin_lock(&new_ui
->ui_lock
);
1580 new_ui
->synced_i_size
= new_ui
->ui_size
;
1581 spin_unlock(&new_ui
->ui_lock
);
1584 * No need to mark whiteout inode clean.
1585 * Whiteout doesn't have non-zero size, no need to update
1586 * synced_i_size for whiteout_ui.
1588 mark_inode_clean(c
, ubifs_inode(old_dir
));
1590 mark_inode_clean(c
, ubifs_inode(new_dir
));
1595 release_head(c
, BASEHD
);
1597 ubifs_ro_mode(c
, err
);
1599 ubifs_delete_orphan(c
, new_inode
->i_ino
);
1601 finish_reservation(c
);
1608 * truncate_data_node - re-compress/encrypt a truncated data node.
1609 * @c: UBIFS file-system description object
1610 * @inode: inode which refers to the data node
1611 * @block: data block number
1612 * @dn: data node to re-compress
1613 * @new_len: new length
1614 * @dn_size: size of the data node @dn in memory
1616 * This function is used when an inode is truncated and the last data node of
1617 * the inode has to be re-compressed/encrypted and re-written.
1619 static int truncate_data_node(const struct ubifs_info
*c
, const struct inode
*inode
,
1620 unsigned int block
, struct ubifs_data_node
*dn
,
1621 int *new_len
, int dn_size
)
1624 int err
, dlen
, compr_type
, out_len
, data_size
;
1626 out_len
= le32_to_cpu(dn
->size
);
1627 buf
= kmalloc_array(out_len
, WORST_COMPR_FACTOR
, GFP_NOFS
);
1631 dlen
= le32_to_cpu(dn
->ch
.len
) - UBIFS_DATA_NODE_SZ
;
1632 data_size
= dn_size
- UBIFS_DATA_NODE_SZ
;
1633 compr_type
= le16_to_cpu(dn
->compr_type
);
1635 if (IS_ENCRYPTED(inode
)) {
1636 err
= ubifs_decrypt(inode
, dn
, &dlen
, block
);
1641 if (compr_type
== UBIFS_COMPR_NONE
) {
1644 err
= ubifs_decompress(c
, &dn
->data
, dlen
, buf
, &out_len
, compr_type
);
1648 ubifs_compress(c
, buf
, *new_len
, &dn
->data
, &out_len
, &compr_type
);
1651 if (IS_ENCRYPTED(inode
)) {
1652 err
= ubifs_encrypt(inode
, dn
, out_len
, &data_size
, block
);
1656 out_len
= data_size
;
1661 ubifs_assert(c
, out_len
<= UBIFS_BLOCK_SIZE
);
1662 dn
->compr_type
= cpu_to_le16(compr_type
);
1663 dn
->size
= cpu_to_le32(*new_len
);
1664 *new_len
= UBIFS_DATA_NODE_SZ
+ out_len
;
1672 * ubifs_jnl_truncate - update the journal for a truncation.
1673 * @c: UBIFS file-system description object
1674 * @inode: inode to truncate
1675 * @old_size: old size
1676 * @new_size: new size
1678 * When the size of a file decreases due to truncation, a truncation node is
1679 * written, the journal tree is updated, and the last data block is re-written
1680 * if it has been affected. The inode is also updated in order to synchronize
1681 * the new inode size.
1683 * This function marks the inode as clean and returns zero on success. In case
1684 * of failure, a negative error code is returned.
1686 int ubifs_jnl_truncate(struct ubifs_info
*c
, const struct inode
*inode
,
1687 loff_t old_size
, loff_t new_size
)
1689 union ubifs_key key
, to_key
;
1690 struct ubifs_ino_node
*ino
;
1691 struct ubifs_trun_node
*trun
;
1692 struct ubifs_data_node
*dn
;
1693 int err
, dlen
, len
, lnum
, offs
, bit
, sz
, sync
= IS_SYNC(inode
);
1695 struct ubifs_inode
*ui
= ubifs_inode(inode
);
1696 ino_t inum
= inode
->i_ino
;
1698 u8 hash_ino
[UBIFS_HASH_ARR_SZ
];
1699 u8 hash_dn
[UBIFS_HASH_ARR_SZ
];
1701 dbg_jnl("ino %lu, size %lld -> %lld",
1702 (unsigned long)inum
, old_size
, new_size
);
1703 ubifs_assert(c
, !ui
->data_len
);
1704 ubifs_assert(c
, S_ISREG(inode
->i_mode
));
1705 ubifs_assert(c
, mutex_is_locked(&ui
->ui_mutex
));
1707 dn_size
= COMPRESSED_DATA_NODE_BUF_SZ
;
1709 if (IS_ENCRYPTED(inode
))
1710 dn_size
+= UBIFS_CIPHER_BLOCK_SIZE
;
1712 sz
= UBIFS_TRUN_NODE_SZ
+ UBIFS_INO_NODE_SZ
+
1713 dn_size
+ ubifs_auth_node_sz(c
);
1715 ino
= kmalloc(sz
, GFP_NOFS
);
1719 trun
= (void *)ino
+ UBIFS_INO_NODE_SZ
;
1720 trun
->ch
.node_type
= UBIFS_TRUN_NODE
;
1721 trun
->inum
= cpu_to_le32(inum
);
1722 trun
->old_size
= cpu_to_le64(old_size
);
1723 trun
->new_size
= cpu_to_le64(new_size
);
1724 zero_trun_node_unused(trun
);
1726 dlen
= new_size
& (UBIFS_BLOCK_SIZE
- 1);
1728 /* Get last data block so it can be truncated */
1729 dn
= (void *)trun
+ UBIFS_TRUN_NODE_SZ
;
1730 blk
= new_size
>> UBIFS_BLOCK_SHIFT
;
1731 data_key_init(c
, &key
, inum
, blk
);
1732 dbg_jnlk(&key
, "last block key ");
1733 err
= ubifs_tnc_lookup(c
, &key
, dn
);
1735 dlen
= 0; /* Not found (so it is a hole) */
1739 int dn_len
= le32_to_cpu(dn
->size
);
1741 if (dn_len
<= 0 || dn_len
> UBIFS_BLOCK_SIZE
) {
1742 ubifs_err(c
, "bad data node (block %u, inode %lu)",
1744 ubifs_dump_node(c
, dn
, dn_size
);
1750 dlen
= 0; /* Nothing to do */
1752 err
= truncate_data_node(c
, inode
, blk
, dn
,
1760 /* Must make reservation before allocating sequence numbers */
1761 len
= UBIFS_TRUN_NODE_SZ
+ UBIFS_INO_NODE_SZ
;
1763 if (ubifs_authenticated(c
))
1764 len
+= ALIGN(dlen
, 8) + ubifs_auth_node_sz(c
);
1768 err
= make_reservation(c
, BASEHD
, len
);
1772 pack_inode(c
, ino
, inode
, 0);
1773 err
= ubifs_node_calc_hash(c
, ino
, hash_ino
);
1777 ubifs_prep_grp_node(c
, trun
, UBIFS_TRUN_NODE_SZ
, dlen
? 0 : 1);
1779 ubifs_prep_grp_node(c
, dn
, dlen
, 1);
1780 err
= ubifs_node_calc_hash(c
, dn
, hash_dn
);
1785 err
= write_head(c
, BASEHD
, ino
, len
, &lnum
, &offs
, sync
);
1789 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
, inum
);
1790 release_head(c
, BASEHD
);
1792 ubifs_add_auth_dirt(c
, lnum
);
1795 sz
= offs
+ UBIFS_INO_NODE_SZ
+ UBIFS_TRUN_NODE_SZ
;
1796 err
= ubifs_tnc_add(c
, &key
, lnum
, sz
, dlen
, hash_dn
);
1801 ino_key_init(c
, &key
, inum
);
1802 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, UBIFS_INO_NODE_SZ
, hash_ino
);
1806 err
= ubifs_add_dirt(c
, lnum
, UBIFS_TRUN_NODE_SZ
);
1810 bit
= new_size
& (UBIFS_BLOCK_SIZE
- 1);
1811 blk
= (new_size
>> UBIFS_BLOCK_SHIFT
) + (bit
? 1 : 0);
1812 data_key_init(c
, &key
, inum
, blk
);
1814 bit
= old_size
& (UBIFS_BLOCK_SIZE
- 1);
1815 blk
= (old_size
>> UBIFS_BLOCK_SHIFT
) - (bit
? 0 : 1);
1816 data_key_init(c
, &to_key
, inum
, blk
);
1818 err
= ubifs_tnc_remove_range(c
, &key
, &to_key
);
1822 finish_reservation(c
);
1823 spin_lock(&ui
->ui_lock
);
1824 ui
->synced_i_size
= ui
->ui_size
;
1825 spin_unlock(&ui
->ui_lock
);
1826 mark_inode_clean(c
, ui
);
1831 release_head(c
, BASEHD
);
1833 ubifs_ro_mode(c
, err
);
1834 finish_reservation(c
);
1842 * ubifs_jnl_delete_xattr - delete an extended attribute.
1843 * @c: UBIFS file-system description object
1845 * @inode: extended attribute inode
1846 * @nm: extended attribute entry name
1848 * This function delete an extended attribute which is very similar to
1849 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1850 * updates the target inode. Returns zero in case of success and a negative
1851 * error code in case of failure.
1853 int ubifs_jnl_delete_xattr(struct ubifs_info
*c
, const struct inode
*host
,
1854 const struct inode
*inode
,
1855 const struct fscrypt_name
*nm
)
1857 int err
, xlen
, hlen
, len
, lnum
, xent_offs
, aligned_xlen
, write_len
;
1858 struct ubifs_dent_node
*xent
;
1859 struct ubifs_ino_node
*ino
;
1860 union ubifs_key xent_key
, key1
, key2
;
1861 int sync
= IS_DIRSYNC(host
);
1862 struct ubifs_inode
*host_ui
= ubifs_inode(host
);
1863 u8 hash
[UBIFS_HASH_ARR_SZ
];
1865 ubifs_assert(c
, inode
->i_nlink
== 0);
1866 ubifs_assert(c
, mutex_is_locked(&host_ui
->ui_mutex
));
1869 * Since we are deleting the inode, we do not bother to attach any data
1870 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1872 xlen
= UBIFS_DENT_NODE_SZ
+ fname_len(nm
) + 1;
1873 aligned_xlen
= ALIGN(xlen
, 8);
1874 hlen
= host_ui
->data_len
+ UBIFS_INO_NODE_SZ
;
1875 len
= aligned_xlen
+ UBIFS_INO_NODE_SZ
+ ALIGN(hlen
, 8);
1877 write_len
= len
+ ubifs_auth_node_sz(c
);
1879 xent
= kzalloc(write_len
, GFP_NOFS
);
1883 /* Make reservation before allocating sequence numbers */
1884 err
= make_reservation(c
, BASEHD
, write_len
);
1890 xent
->ch
.node_type
= UBIFS_XENT_NODE
;
1891 xent_key_init(c
, &xent_key
, host
->i_ino
, nm
);
1892 key_write(c
, &xent_key
, xent
->key
);
1894 xent
->type
= get_dent_type(inode
->i_mode
);
1895 xent
->nlen
= cpu_to_le16(fname_len(nm
));
1896 memcpy(xent
->name
, fname_name(nm
), fname_len(nm
));
1897 xent
->name
[fname_len(nm
)] = '\0';
1898 zero_dent_node_unused(xent
);
1899 ubifs_prep_grp_node(c
, xent
, xlen
, 0);
1901 ino
= (void *)xent
+ aligned_xlen
;
1902 pack_inode(c
, ino
, inode
, 0);
1903 ino
= (void *)ino
+ UBIFS_INO_NODE_SZ
;
1904 pack_inode(c
, ino
, host
, 1);
1905 err
= ubifs_node_calc_hash(c
, ino
, hash
);
1909 err
= write_head(c
, BASEHD
, xent
, write_len
, &lnum
, &xent_offs
, sync
);
1911 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
, host
->i_ino
);
1912 release_head(c
, BASEHD
);
1914 ubifs_add_auth_dirt(c
, lnum
);
1919 /* Remove the extended attribute entry from TNC */
1920 err
= ubifs_tnc_remove_nm(c
, &xent_key
, nm
);
1923 err
= ubifs_add_dirt(c
, lnum
, xlen
);
1928 * Remove all nodes belonging to the extended attribute inode from TNC.
1929 * Well, there actually must be only one node - the inode itself.
1931 lowest_ino_key(c
, &key1
, inode
->i_ino
);
1932 highest_ino_key(c
, &key2
, inode
->i_ino
);
1933 err
= ubifs_tnc_remove_range(c
, &key1
, &key2
);
1936 err
= ubifs_add_dirt(c
, lnum
, UBIFS_INO_NODE_SZ
);
1940 /* And update TNC with the new host inode position */
1941 ino_key_init(c
, &key1
, host
->i_ino
);
1942 err
= ubifs_tnc_add(c
, &key1
, lnum
, xent_offs
+ len
- hlen
, hlen
, hash
);
1946 finish_reservation(c
);
1947 spin_lock(&host_ui
->ui_lock
);
1948 host_ui
->synced_i_size
= host_ui
->ui_size
;
1949 spin_unlock(&host_ui
->ui_lock
);
1950 mark_inode_clean(c
, host_ui
);
1955 release_head(c
, BASEHD
);
1957 ubifs_ro_mode(c
, err
);
1958 finish_reservation(c
);
1963 * ubifs_jnl_change_xattr - change an extended attribute.
1964 * @c: UBIFS file-system description object
1965 * @inode: extended attribute inode
1968 * This function writes the updated version of an extended attribute inode and
1969 * the host inode to the journal (to the base head). The host inode is written
1970 * after the extended attribute inode in order to guarantee that the extended
1971 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1972 * consequently, the write-buffer is synchronized. This function returns zero
1973 * in case of success and a negative error code in case of failure.
1975 int ubifs_jnl_change_xattr(struct ubifs_info
*c
, const struct inode
*inode
,
1976 const struct inode
*host
)
1978 int err
, len1
, len2
, aligned_len
, aligned_len1
, lnum
, offs
;
1979 struct ubifs_inode
*host_ui
= ubifs_inode(host
);
1980 struct ubifs_ino_node
*ino
;
1981 union ubifs_key key
;
1982 int sync
= IS_DIRSYNC(host
);
1983 u8 hash_host
[UBIFS_HASH_ARR_SZ
];
1984 u8 hash
[UBIFS_HASH_ARR_SZ
];
1986 dbg_jnl("ino %lu, ino %lu", host
->i_ino
, inode
->i_ino
);
1987 ubifs_assert(c
, inode
->i_nlink
> 0);
1988 ubifs_assert(c
, mutex_is_locked(&host_ui
->ui_mutex
));
1990 len1
= UBIFS_INO_NODE_SZ
+ host_ui
->data_len
;
1991 len2
= UBIFS_INO_NODE_SZ
+ ubifs_inode(inode
)->data_len
;
1992 aligned_len1
= ALIGN(len1
, 8);
1993 aligned_len
= aligned_len1
+ ALIGN(len2
, 8);
1995 aligned_len
+= ubifs_auth_node_sz(c
);
1997 ino
= kzalloc(aligned_len
, GFP_NOFS
);
2001 /* Make reservation before allocating sequence numbers */
2002 err
= make_reservation(c
, BASEHD
, aligned_len
);
2006 pack_inode(c
, ino
, host
, 0);
2007 err
= ubifs_node_calc_hash(c
, ino
, hash_host
);
2010 pack_inode(c
, (void *)ino
+ aligned_len1
, inode
, 1);
2011 err
= ubifs_node_calc_hash(c
, (void *)ino
+ aligned_len1
, hash
);
2015 err
= write_head(c
, BASEHD
, ino
, aligned_len
, &lnum
, &offs
, 0);
2016 if (!sync
&& !err
) {
2017 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
2019 ubifs_wbuf_add_ino_nolock(wbuf
, host
->i_ino
);
2020 ubifs_wbuf_add_ino_nolock(wbuf
, inode
->i_ino
);
2022 release_head(c
, BASEHD
);
2026 ubifs_add_auth_dirt(c
, lnum
);
2028 ino_key_init(c
, &key
, host
->i_ino
);
2029 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, len1
, hash_host
);
2033 ino_key_init(c
, &key
, inode
->i_ino
);
2034 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
+ aligned_len1
, len2
, hash
);
2038 finish_reservation(c
);
2039 spin_lock(&host_ui
->ui_lock
);
2040 host_ui
->synced_i_size
= host_ui
->ui_size
;
2041 spin_unlock(&host_ui
->ui_lock
);
2042 mark_inode_clean(c
, host_ui
);
2047 release_head(c
, BASEHD
);
2049 ubifs_ro_mode(c
, err
);
2050 finish_reservation(c
);